VGG 16 with extra layers

Accuracy:

Training 86% Validation 84%

Loss:

Training 0.4118 Validation 0.5059

In [10]:
import warnings
warnings.filterwarnings("ignore")
In [11]:
tags = { "C0": "safe driving",
"C1": "texting - right",
"C2": "talking on the phone - right",
"C3": "texting - left",
"C4": "talking on the phone - left",
"C5": "operating the radio",
"C6": "drinking",
"C7": "reaching behind",
"C8": "hair and makeup",
"C9": "talking to passenger" }
In [12]:
import os
import random
import numpy as np
import pandas as pd 
from IPython.display import display
import matplotlib.pyplot as plt
import cv2

from tensorflow.keras.models import Model
from tensorflow.keras.layers import GlobalAveragePooling2D, Dense

import tensorflow.keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten
from tensorflow.keras.layers import Conv2D, MaxPooling2D
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.preprocessing import image 
from tensorflow.keras.layers import BatchNormalization
from tensorflow.keras import optimizers
from tensorflow.keras.models import model_from_json
In [13]:
test_image = []
i = 0
fig, ax = plt.subplots(1, 20, figsize = (50,50 ))

files = os.listdir('../../statefarm_dataset/test')
nums = np.random.randint(low=1, high=len(files), size=20)
for i in range(20):
#     print ('Image number:',i)
    img = cv2.imread('../../statefarm_dataset/test/'+files[nums[i]])
    #img = color.rgb2gray(img)
    img = img[50:,120:-50]
    img = cv2.resize(img,(224,224))
    test_image.append(img)
    ax[i].imshow(img,cmap = 'gray')
    plt.show
In [14]:
with open('model_vgg16_all_layers.json','r') as f:
    model = model_from_json(f.read())


# Compile CNN model
sgd = optimizers.SGD(lr = 0.001)
model.compile(loss='categorical_crossentropy',optimizer = sgd,metrics=['accuracy'])
print(model.summary())

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
Image_input (InputLayer)     [(None, 224, 224, 3)]     0         
_________________________________________________________________
vgg16 (Model)                (None, 7, 7, 512)         14714688  
_________________________________________________________________
global_average_pooling2d_2 ( (None, 512)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 1024)              525312    
_________________________________________________________________
dropout_4 (Dropout)          (None, 1024)              0         
_________________________________________________________________
dense_7 (Dense)              (None, 1024)              1049600   
_________________________________________________________________
batch_normalization_2 (Batch (None, 1024)              4096      
_________________________________________________________________
dropout_5 (Dropout)          (None, 1024)              0         
_________________________________________________________________
dense_8 (Dense)              (None, 512)               524800    
_________________________________________________________________
predictions (Dense)          (None, 10)                5130      
=================================================================
Total params: 16,823,626
Trainable params: 16,821,578
Non-trainable params: 2,048
_________________________________________________________________
None
In [15]:
model.load_weights('model_vgg16_all_layers.h5')

test = []

for img in test_image:
    test.append(img)

predict_test = np.array(test).reshape(-1,224,224,3).astype('float32')
prediction = model.predict(predict_test)
print(prediction[0])

i = 0
fig, ax = plt.subplots(20, 1, figsize = (100,100))

for i in range(20):
    ax[i].imshow(test[i])
    predicted_class = 'C'+str(np.where(prediction[i] == np.amax(prediction[i]))[0][0])
    ax[i].set_title(tags[predicted_class])
    plt.show
    

[7.0517140e-06 2.1268957e-11 1.2101271e-11 1.0572809e-11 1.3548809e-10
 4.8984643e-09 6.9798611e-12 1.8341987e-07 1.5569078e-05 9.9997711e-01]

MobileNet Model

Accuracy:

Training 90% Validation 83%

Loss:

Training 0.30 Validation 0.49

In [16]:
with open('mobilenet_sgd_nolayers.json','r') as f:
    mobile_model = model_from_json(f.read())


# Compile CNN model
sgd = optimizers.SGD(lr = 0.001)
mobile_model.compile(loss='categorical_crossentropy',optimizer = sgd,metrics=['accuracy'])
print(mobile_model.summary())

Model: "model_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         [(None, None, None, 3)]   0         
_________________________________________________________________
conv1_pad (ZeroPadding2D)    (None, None, None, 3)     0         
_________________________________________________________________
conv1 (Conv2D)               (None, None, None, 32)    864       
_________________________________________________________________
conv1_bn (BatchNormalization (None, None, None, 32)    128       
_________________________________________________________________
conv1_relu (ReLU)            (None, None, None, 32)    0         
_________________________________________________________________
conv_dw_1 (DepthwiseConv2D)  (None, None, None, 32)    288       
_________________________________________________________________
conv_dw_1_bn (BatchNormaliza (None, None, None, 32)    128       
_________________________________________________________________
conv_dw_1_relu (ReLU)        (None, None, None, 32)    0         
_________________________________________________________________
conv_pw_1 (Conv2D)           (None, None, None, 64)    2048      
_________________________________________________________________
conv_pw_1_bn (BatchNormaliza (None, None, None, 64)    256       
_________________________________________________________________
conv_pw_1_relu (ReLU)        (None, None, None, 64)    0         
_________________________________________________________________
conv_pad_2 (ZeroPadding2D)   (None, None, None, 64)    0         
_________________________________________________________________
conv_dw_2 (DepthwiseConv2D)  (None, None, None, 64)    576       
_________________________________________________________________
conv_dw_2_bn (BatchNormaliza (None, None, None, 64)    256       
_________________________________________________________________
conv_dw_2_relu (ReLU)        (None, None, None, 64)    0         
_________________________________________________________________
conv_pw_2 (Conv2D)           (None, None, None, 128)   8192      
_________________________________________________________________
conv_pw_2_bn (BatchNormaliza (None, None, None, 128)   512       
_________________________________________________________________
conv_pw_2_relu (ReLU)        (None, None, None, 128)   0         
_________________________________________________________________
conv_dw_3 (DepthwiseConv2D)  (None, None, None, 128)   1152      
_________________________________________________________________
conv_dw_3_bn (BatchNormaliza (None, None, None, 128)   512       
_________________________________________________________________
conv_dw_3_relu (ReLU)        (None, None, None, 128)   0         
_________________________________________________________________
conv_pw_3 (Conv2D)           (None, None, None, 128)   16384     
_________________________________________________________________
conv_pw_3_bn (BatchNormaliza (None, None, None, 128)   512       
_________________________________________________________________
conv_pw_3_relu (ReLU)        (None, None, None, 128)   0         
_________________________________________________________________
conv_pad_4 (ZeroPadding2D)   (None, None, None, 128)   0         
_________________________________________________________________
conv_dw_4 (DepthwiseConv2D)  (None, None, None, 128)   1152      
_________________________________________________________________
conv_dw_4_bn (BatchNormaliza (None, None, None, 128)   512       
_________________________________________________________________
conv_dw_4_relu (ReLU)        (None, None, None, 128)   0         
_________________________________________________________________
conv_pw_4 (Conv2D)           (None, None, None, 256)   32768     
_________________________________________________________________
conv_pw_4_bn (BatchNormaliza (None, None, None, 256)   1024      
_________________________________________________________________
conv_pw_4_relu (ReLU)        (None, None, None, 256)   0         
_________________________________________________________________
conv_dw_5 (DepthwiseConv2D)  (None, None, None, 256)   2304      
_________________________________________________________________
conv_dw_5_bn (BatchNormaliza (None, None, None, 256)   1024      
_________________________________________________________________
conv_dw_5_relu (ReLU)        (None, None, None, 256)   0         
_________________________________________________________________
conv_pw_5 (Conv2D)           (None, None, None, 256)   65536     
_________________________________________________________________
conv_pw_5_bn (BatchNormaliza (None, None, None, 256)   1024      
_________________________________________________________________
conv_pw_5_relu (ReLU)        (None, None, None, 256)   0         
_________________________________________________________________
conv_pad_6 (ZeroPadding2D)   (None, None, None, 256)   0         
_________________________________________________________________
conv_dw_6 (DepthwiseConv2D)  (None, None, None, 256)   2304      
_________________________________________________________________
conv_dw_6_bn (BatchNormaliza (None, None, None, 256)   1024      
_________________________________________________________________
conv_dw_6_relu (ReLU)        (None, None, None, 256)   0         
_________________________________________________________________
conv_pw_6 (Conv2D)           (None, None, None, 512)   131072    
_________________________________________________________________
conv_pw_6_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_6_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_7 (DepthwiseConv2D)  (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_7_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_7_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_7 (Conv2D)           (None, None, None, 512)   262144    
_________________________________________________________________
conv_pw_7_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_7_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_8 (DepthwiseConv2D)  (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_8_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_8_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_8 (Conv2D)           (None, None, None, 512)   262144    
_________________________________________________________________
conv_pw_8_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_8_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_9 (DepthwiseConv2D)  (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_9_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_9_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_9 (Conv2D)           (None, None, None, 512)   262144    
_________________________________________________________________
conv_pw_9_bn (BatchNormaliza (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_9_relu (ReLU)        (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_10 (DepthwiseConv2D) (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_10_bn (BatchNormaliz (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_10_relu (ReLU)       (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_10 (Conv2D)          (None, None, None, 512)   262144    
_________________________________________________________________
conv_pw_10_bn (BatchNormaliz (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_10_relu (ReLU)       (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_11 (DepthwiseConv2D) (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_11_bn (BatchNormaliz (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_11_relu (ReLU)       (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_11 (Conv2D)          (None, None, None, 512)   262144    
_________________________________________________________________
conv_pw_11_bn (BatchNormaliz (None, None, None, 512)   2048      
_________________________________________________________________
conv_pw_11_relu (ReLU)       (None, None, None, 512)   0         
_________________________________________________________________
conv_pad_12 (ZeroPadding2D)  (None, None, None, 512)   0         
_________________________________________________________________
conv_dw_12 (DepthwiseConv2D) (None, None, None, 512)   4608      
_________________________________________________________________
conv_dw_12_bn (BatchNormaliz (None, None, None, 512)   2048      
_________________________________________________________________
conv_dw_12_relu (ReLU)       (None, None, None, 512)   0         
_________________________________________________________________
conv_pw_12 (Conv2D)          (None, None, None, 1024)  524288    
_________________________________________________________________
conv_pw_12_bn (BatchNormaliz (None, None, None, 1024)  4096      
_________________________________________________________________
conv_pw_12_relu (ReLU)       (None, None, None, 1024)  0         
_________________________________________________________________
conv_dw_13 (DepthwiseConv2D) (None, None, None, 1024)  9216      
_________________________________________________________________
conv_dw_13_bn (BatchNormaliz (None, None, None, 1024)  4096      
_________________________________________________________________
conv_dw_13_relu (ReLU)       (None, None, None, 1024)  0         
_________________________________________________________________
conv_pw_13 (Conv2D)          (None, None, None, 1024)  1048576   
_________________________________________________________________
conv_pw_13_bn (BatchNormaliz (None, None, None, 1024)  4096      
_________________________________________________________________
conv_pw_13_relu (ReLU)       (None, None, None, 1024)  0         
_________________________________________________________________
global_average_pooling2d_2 ( (None, 1024)              0         
_________________________________________________________________
dense_4 (Dense)              (None, 10)                10250     
=================================================================
Total params: 3,239,114
Trainable params: 3,217,226
Non-trainable params: 21,888
_________________________________________________________________
None
In [17]:
mobile_model.load_weights('mobilenet_sgd_nolayerss.h5')

test = []

for img in test_image:
    test.append(img)

predict_test = np.array(test).reshape(-1,224,224,3).astype('float32')
prediction = mobile_model.predict(predict_test)
print(prediction[0])

i = 0
fig, ax = plt.subplots(20, 1, figsize = (100,100))

for i in range(20):
    ax[i].imshow(test[i])
    predicted_class = 'C'+str(np.where(prediction[i] == np.amax(prediction[i]))[0][0])
    ax[i].set_title(tags[predicted_class])
    plt.show
    

[5.3949364e-02 1.3800645e-08 7.3670008e-09 3.7311833e-07 2.1908625e-06
 8.8585314e-04 1.7493765e-08 1.0414767e-05 1.2859858e-03 9.4386578e-01]